Pretty much what dndw says. It's possible to make the AI play optimally, but in reality, it'll probably be a pretty boring AI.
Also, machine learned AIs usually zones into a repeatable local maxima, where they do stuff like spam low kick repeatedly in fighting games.
I already overhauled the AI once when doing my mod. The original AI was even more random than the current one and pretty much only uses a few skills. Right now, there is a bug with the AI, and it's causing certain situations to come up much more often than it should. I'll fix this some time, but debugging the AI is a pain in the ass, and I really don't have the motivation to do it right now (like I said, kind of busy lately).
If I wanted to upgrade the AI, I do agree that a few turn lookahead would make the AI a lot more interesting. The main problem with that is performance. For better or for worse, the code is what it is, and in order to do 2-3 turn lookaheads with the current codebase, I'm willing to bet it'll take 3-5 seconds per turn for the AI to decide what to do. Right now, each skills is non-deterministic, and doing a min-max search of the game tree would be pretty slow, especially with the current implementation of the game state where the entire combat is cloned every step.
I'm willing to bet that a monte carlo tree search would give better performance and results, but it still really boils down to writing better heuristics.
ANYWAYS, I guess what I'm saying is this mod is a hobby, and I can't be arsed to be do a huge game mechanic overhaul right now. Especially if it's actually non-trivial
It's supposed to be a break from work, not work itself!
As for difficulty, I'm not really sure what to do here. Honestly the AI weights I thought was supposed to provide flavor, not difficulty. The AI is going to do what it does, but I guess part of the problem with the AI now is that it doesn't really take willpower into account as much as it should. Also, it doesn't check the end turn results. I'll try to work that into the next build so at the very least the AI favors lowering your will power a bit more. However I'm a bit afraid that it will just disregard arousal. Ergh.
There will probably always be a few player exploits that you can do that will trivialize the AI. This happens in a lot of games to be honest, so I'm not too upset about it. That said, spiral thrust is pretty silly right now, what would you say is a good way to nerf it further? I don't want to hurt the early game potential too hard, but the late game pleasure damage is a bit ridiculous. For now I think I'll make the lethargic debuff you get from it last 30 turns, and make it last outside of combat, so there's a real downside for using it as a certain kill move.
Edit: I did fix the hypnotic semen bug though just now.